Add paper and GitHub links to dataset card

README.md

@@ -2,8 +2,12 @@
 language:
 - en
 license: cc-by-4.0
-license_link: https://huggingface.co/datasets/logo-lab/trl-rbench/blob/main/LICENSES.md
+size_categories:
+- 1M<n<10M
 pretty_name: TRL-Rbench
+task_categories:
+- other
+license_link: https://huggingface.co/datasets/logo-lab/trl-rbench/blob/main/LICENSES.md
 tags:
 - tabular
 - row-level
@@ -12,24 +16,32 @@ tags:
 - entity-matching
 - record-linkage
 - trl-bench
-size_categories:
-- 1M<n<10M
 configs:
 - config_name: row_prediction
   data_files:
-  - {split: train,      path: data/row_prediction/train-*.parquet}
-  - {split: validation, path: data/row_prediction/validation-*.parquet}
-  - {split: test,       path: data/row_prediction/test-*.parquet}
+  - split: train
+    path: data/row_prediction/train-*.parquet
+  - split: validation
+    path: data/row_prediction/validation-*.parquet
+  - split: test
+    path: data/row_prediction/test-*.parquet
 - config_name: record_linkage
   data_files:
-  - {split: train,      path: data/record_linkage/train-*.parquet}
-  - {split: validation, path: data/record_linkage/validation-*.parquet}
-  - {split: test,       path: data/record_linkage/test-*.parquet}
+  - split: train
+    path: data/record_linkage/train-*.parquet
+  - split: validation
+    path: data/record_linkage/validation-*.parquet
+  - split: test
+    path: data/record_linkage/test-*.parquet
 ---
 
 # TRL-Rbench
 
-Row-level evaluation suite of TRL-Bench:
+This is the row-level evaluation suite of **TRL-Bench**, as presented in the paper [TRL-Bench: Standardizing Cross-Paradigm Representation-Level Evaluation of Tabular Encoders](https://huggingface.co/papers/2606.09323).
+
+**Official Code:** [https://github.com/LOGO-CUHKSZ/TRL-Bench](https://github.com/LOGO-CUHKSZ/TRL-Bench)
+
+The suite consists of:
 
 - **`row_prediction`**: 50 OpenML tables, 1.1M rows (887,720 train / 110,962 validation / 110,983 test) across 123 hand-verified targets. Per-row schema bundles features and targets as JSON dicts so all 50 tables share one config.
 - **`record_linkage`**: 16 entity-matching sources (8 clean DeepMatcher + 4 dirty DeepMatcher + 4 WDC LSPM v2 sizes) unified into one config with 357,833 train / 95,817 validation / 42,835 test row pairs. Per-row stores both records as JSON dicts plus `source` / `family` / `pair_id` / `label`.
@@ -100,4 +112,4 @@ print(json.loads(sample["table_b_record_json"]))
 rl_beer = rl.filter(lambda x: x["source"] == "deepmatcher_beer")
 ```
 
-See `LICENSES.md` for per-source license details.
+See `LICENSES.md` for per-source license details.